libstdc++-v3/include/parallel/par_loop.h

   1 // -*- C++ -*-
   2
   3 // Copyright (C) 2007, 2008, 2009 Free Software Foundation, Inc.
   4 //
   5 // This file is part of the GNU ISO C++ Library.  This library is free
   6 // software; you can redistribute it and/or modify it under the terms
   7 // of the GNU General Public License as published by the Free Software
   8 // Foundation; either version 3, or (at your option) any later
   9 // version.
  10
  11 // This library is distributed in the hope that it will be useful, but
  12 // WITHOUT ANY WARRANTY; without even the implied warranty of
  13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 // General Public License for more details.
  15
  16 // Under Section 7 of GPL version 3, you are granted additional
  17 // permissions described in the GCC Runtime Library Exception, version
  18 // 3.1, as published by the Free Software Foundation.
  19
  20 // You should have received a copy of the GNU General Public License and
  21 // a copy of the GCC Runtime Library Exception along with this program;
  22 // see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  23 // <http://www.gnu.org/licenses/>.
  24
  25 /** @file parallel/par_loop.h
  26  *  @brief Parallelization of embarrassingly parallel execution by
  27  *  means of equal splitting.
  28  *  This file is a GNU parallel extension to the Standard C++ Library.
  29  */
  30
  31 // Written by Felix Putze.
  32
  33 #ifndef _GLIBCXX_PARALLEL_PAR_LOOP_H
  34 #define _GLIBCXX_PARALLEL_PAR_LOOP_H 1
  35
  36 #include <omp.h>
  37 #include <parallel/settings.h>
  38 #include <parallel/base.h>
  39 #include <parallel/equally_split.h>
  40
  41 namespace __gnu_parallel
  42 {
  43
  44 /** @brief Embarrassingly parallel algorithm for random access
  45   * iterators, using hand-crafted parallelization by equal splitting
  46   * the work.
  47   *
  48   *  @param begin Begin iterator of element sequence.
  49   *  @param end End iterator of element sequence.
  50   *  @param o User-supplied functor (comparator, predicate, adding
  51   *  functor, ...)
  52   *  @param f Functor to "process" an element with op (depends on
  53   *  desired functionality, e. g. for std::for_each(), ...).
  54   *  @param r Functor to "add" a single result to the already
  55   *  processed elements (depends on functionality).
  56   *  @param base Base value for reduction.
  57   *  @param output Pointer to position where final result is written to
  58   *  @param bound Maximum number of elements processed (e. g. for
  59   *  std::count_n()).
  60   *  @return User-supplied functor (that may contain a part of the result).
  61   */
  62 template<typename RandomAccessIterator,
  63          typename Op,
  64          typename Fu,
  65          typename Red,
  66          typename Result>
  67   Op
  68   for_each_template_random_access_ed(RandomAccessIterator begin,
  69                                      RandomAccessIterator end,
  70                                      Op o, Fu& f, Red r, Result base,
  71                                      Result& output,
  72                                      typename std::iterator_traits
  73                                      <RandomAccessIterator>::
  74                                      difference_type bound)
  75   {
  76     typedef std::iterator_traits<RandomAccessIterator> traits_type;
  77     typedef typename traits_type::difference_type difference_type;
  78     const difference_type length = end - begin;
  79     Result *thread_results;
  80     bool* constructed;
  81
  82     thread_index_t num_threads =
  83       __gnu_parallel::min<difference_type>(get_max_threads(), length);
  84
  85 #   pragma omp parallel num_threads(num_threads)
  86       {
  87 #       pragma omp single
  88           {
  89             num_threads = omp_get_num_threads();
  90             thread_results = static_cast<Result*>(
  91                                 ::operator new(num_threads * sizeof(Result)));
  92             constructed = new bool[num_threads];
  93           }
  94
  95         thread_index_t iam = omp_get_thread_num();
  96
  97         // Neutral element.
  98         Result* reduct = static_cast<Result*>(::operator new(sizeof(Result)));
  99
 100         difference_type
 101             start = equally_split_point(length, num_threads, iam),
 102             stop = equally_split_point(length, num_threads, iam + 1);
 103
 104         if (start < stop)
 105           {
 106             new(reduct) Result(f(o, begin + start));
 107             ++start;
 108             constructed[iam] = true;
 109           }
 110         else
 111           constructed[iam] = false;
 112
 113         for (; start < stop; ++start)
 114           *reduct = r(*reduct, f(o, begin + start));
 115
 116         thread_results[iam] = *reduct;
 117       } //parallel
 118
 119     for (thread_index_t i = 0; i < num_threads; ++i)
 120         if (constructed[i])
 121             output = r(output, thread_results[i]);
 122
 123     // Points to last element processed (needed as return value for
 124     // some algorithms like transform).
 125     f.finish_iterator = begin + length;
 126
 127     delete[] thread_results;
 128     delete[] constructed;
 129
 130     return o;
 131   }
 132
 133 } // end namespace
 134
 135 #endif /* _GLIBCXX_PARALLEL_PAR_LOOP_H */